agent-intel

name	agent-intel
description	Deep on-chain intelligence analysis of hackathon trading agents - validation attestations, reputation feedback, trade intents, sybil detection, and scoring breakdown.

Agent Intel - On-Chain Agent Analysis

Analyze any agent registered on the hackathon's ERC-8004 contracts. Produces a full intelligence report covering registration, trading, validation, reputation, and gaming detection.

Prerequisites

• .env file with CHAIN_RPC_URL and HACKATHON_* addresses
• cast (foundry) installed
• Go toolchain (for leaderboard-spy)

Workflow

Step 1: Identify the target

If user provides a name, first run the agent list to find the ID:

HACKATHON_VALIDATION_ADDR=0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1 \
  bash -c 'set -a && . ./.env && set +a && \
  export HACKATHON_VALIDATION_ADDR=0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1 && \
  cd golang && go run ./cmd/leaderboard-spy agents'

Find the agent by name in the output. Note its ID (first column).

If user provides an ID or wallet address, use that directly.

Step 2: Collect on-chain data

Run these queries in parallel using the RPC from .env (CHAIN_RPC_URL). All are read-only (no gas).

Contract addresses (hackathon):

• AgentRegistry: 0x97b07dDc405B0c28B17559aFFE63BdB3632d0ca3
• ValidationRegistry: 0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1
• ReputationRegistry: 0x423a9904e39537a9997fbaF0f220d79D7d545763
• RiskRouter: 0xd6A6952545FF6E6E6681c2d15C59f9EB8F40FdBC
• HackathonVault: 0x0E7CD8ef9743FEcf94f9103033a044caBD45fC90

Agent ID as hex topic (pad to 32 bytes): e.g., agent 18 = 0x0000000000000000000000000000000000000000000000000000000000000012

2a. Basic info

# Wallet nonce and balance
cast nonce <wallet> --rpc-url $RPC
cast balance <wallet> --ether --rpc-url $RPC

2b. Attestation notes (via leaderboard-spy)

HACKATHON_VALIDATION_ADDR=0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1 \
  bash -c 'set -a && . ./.env && set +a && \
  export HACKATHON_VALIDATION_ADDR=0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1 && \
  cd golang && go run ./cmd/leaderboard-spy attestations <AGENT_ID>'

2c. Validation events (from ValidationRegistry)

Event signature: 0x5c19b748ded05e13d6fb5776172b6f813692e237572a5c725a43077cb82a67db

cast logs --from-block <10580000> --to-block <LATEST> \
  --address 0x92bF63E5C7Ac6980f237a7164Ab413BE226187F1 \
  "0x5c19b748ded05e13d6fb5776172b6f813692e237572a5c725a43077cb82a67db" \
  "<AGENT_ID_HEX_PADDED>" \
  --rpc-url $RPC > /tmp/agent_val.txt

Analyze: count events, extract unique validator addresses from topic[2], check if all from same address (self-attestation).

2d. Reputation events (from ReputationRegistry)

Event signature: 0x2957247c48ca9e733e8c76cff7e64dafa0f9402a35dbe2cb1cfc7259077435e4

cast logs --from-block <10580000> --to-block <LATEST> \
  --address 0x423a9904e39537a9997fbaF0f220d79D7d545763 \
  "0x2957247c48ca9e733e8c76cff7e64dafa0f9402a35dbe2cb1cfc7259077435e4" \
  "<AGENT_ID_HEX_PADDED>" \
  --rpc-url $RPC > /tmp/agent_rep.txt

Analyze with Python:

import re
from collections import Counter

with open('/tmp/agent_rep.txt') as f:
    content = f.read()

events = content.split('- address:')
events = [e for e in events if e.strip()]

validators = {}
scores = []
for event in events:
    topics = re.findall(r'0x[0-9a-fA-F]{64}', event.split('topics:')[1].split(']')[0]) if 'topics:' in event else []
    if len(topics) >= 3:
        validator = '0x' + topics[2][26:]
        validators[validator] = validators.get(validator, 0) + 1
    data_match = re.search(r'data: (0x[0-9a-fA-F]+)', event)
    if data_match:
        data = data_match.group(1)[2:]
        score = int(data[:64], 16)
        scores.append(score)

print(f"Total: {len(events)}, Unique validators: {len(validators)}")
print(f"Score distribution: {dict(Counter(scores).most_common())}")
# Sybil indicator: unique_validators == total_events AND all nonce=1

2e. Trade intents (from RiskRouter)

Event signature: 0x536c9b7dd53ffa0a0b01880535f363a405c6a20ebedc6802702927c602852b9b

cast logs --from-block <10580000> --to-block <LATEST> \
  --address 0xd6A6952545FF6E6E6681c2d15C59f9EB8F40FdBC \
  "0x536c9b7dd53ffa0a0b01880535f363a405c6a20ebedc6802702927c602852b9b" \
  "<AGENT_ID_HEX_PADDED>" \
  --rpc-url $RPC > /tmp/agent_trades.txt

Decode trade data with Python (ABI-encoded: pair string, action string, amountUsdScaled uint256):

for event in events:
    data = re.search(r'data: (0x[0-9a-fA-F]+)', event).group(1)[2:]
    amount_scaled = int(data[128:192], 16)
    amount_usd = amount_scaled / 100.0
    pair_offset = int(data[0:64], 16) * 2
    pair_len = int(data[pair_offset:pair_offset+64], 16)
    pair = bytes.fromhex(data[pair_offset+64:pair_offset+64+pair_len*2]).decode()
    action_offset = int(data[64:128], 16) * 2
    action_len = int(data[action_offset:action_offset+64], 16)
    action = bytes.fromhex(data[action_offset+64:action_offset+64+action_len*2]).decode()

Step 3: Analyze and detect gaming

Check these indicators:

Signal	What to check
Self-attestation	All validation attestations posted by the agent's own wallet (or a single address). Check if unique validator count == 1 for the agent's attestations. The validator address in the event is whoever called postEIP712Attestation - if it's always the same address, the agent is self-attesting.
Attestation spam	>100 attestations, notes containing "boost"
Reputation sybil	Many feedbacks, each from unique address with nonce=1
Wash trading	Many small trades ($1-$10), repetitive pairs
Score inflation	Scores consistently 95-100 across all feedbacks

Step 4: Produce the report

Structure the output as:

## [Agent Name] (Agent #ID) - Intelligence Report

### Registration
- Wallet, operator, description, capabilities
- Wallet nonce and balance (activity indicator)
- Vault status

### Leaderboard Scores
- Reputation: X, Validation: X, Trades: X

### Trading Activity
- Total intents, pairs traded, actions, volume
- Average trade size, time range

### Validation Attestations
- Count, unique validators, score distribution
- Self-attestation detected? Boost entries?

### Reputation Feedback
- Count, unique validators, score distribution
- Sybil indicators (unique addresses, nonce=1 pattern)

### Gaming Assessment
- Summary of detected gaming techniques
- Comparison to baseline (our agents)

Scoring Formula Reference

The leaderboard reputation score is calculated by a judge bot (runs every ~4h):

• Validation avg x 0.50 = 0-50 pts
• Trade count x 3 (capped at 10 trades = 30 pts)
• Vault claimed = 10 pts
• Any attestation posted = 10 pts
• Max: 100

The getAverageScore on ReputationRegistry returns the average of all giveFeedback values posted by external wallets. This is separate from the judge bot score but displayed as "Reputation" on the leaderboard.

Our Agents for Comparison

Agent	ID	Reputation	Validation	Attestations	Trades
Swiftward Alpha	32	70	100	5	2
SWIFTWARD_BETA	42	0	0	0	0
Swiftward Gamma	43	31	0	0	1
Random Trader	37	77	100	613	5